I suffer from several biases. Some of them are probably rational, at least, thatâs what I tell myself. Itâs a bias in its own right. But, I think Iâve been conditioned to digest a future Coda/Superhuman roadmap claims with an element of skepticism that is reinforced by market and technical realities. I have some work I must do to temper my skepticism and remain as open-minded as possible. I truly hope the agent builder isnât another Coda Brain. So far, this vision is as elusive as the brain.
Still, itâs a vision I hope for because it bridges a critical gap that enterprise-centric solutions require - security, determinism, and provability. Despite Anthropicâs deep enterprise penetration, it was not achieved by Claude Code or Claude CoWork. These tools run in the faceâcompletely antitheticalâof the primary business drivers that helped Anthropic rise to enterprise dominance.
Superhuman magically bridges this gap with its presence in the OS and its predilection for cloud integration, and soon, with Packs. Itâs still unclear to me if something this simple will work in a Pack. It should, right? Thoughts?
// This is the "sprinkle of code" that probably matters most
pack.addMCPServer({
name: "LocalFileSystem",
endpointUrl: "http://localhost:3000",
});
Ultimately, natural language and other human-guided intentions is how we will instruct computers. This is rarely a debate narrative, and when it does arise, itâs a waste of time and energy. We want our computers to understand us in the biological framework we exist. Itâs not unlike self-driving cars - they need to operate on roads that people designed for people.
Your examples of pseudo code as prompts are very good. They exist in the grey area of discerning user intent and serve as crucial language examples that LLMs seem to prefer. Perhaps a sidebar observation, Iâve been experimenting with real-time voice as the basis for pseudo code generation. I say what I want the process to do, and it writes the workflow in a yaml/code-like format ready to run in a skill. Note - this all occurred in a Coda document. It can occur in any app because VoiceInk, like Superhuman GO, are OS-bound hooks.
Ideally, no human should have to write these code-like constructs and todayâs LLMs are capable of abstracting instructions in the LLMs native tongue in real-time. Itâs good experience to do this manually, but you can also eliminate it entirely by using tools like VoiceInk which provide the design components to accelerate this part of the process.
Yes, but with shields up, perhaps? 
â¤đđđ
